Algorithmic Pricing: How Amazon, Uber and Delta Set Your Price, and What the Law Now Says

By Tom Morgan · Updated October 26, 2026 ·

Quick answer

• Amazon is not mainly a “different price for you” story. The FTC alleges its algorithms pushed prices up across whole markets. The FTC’s antitrust trial is set for March 29, 2027, after the judge refused Amazon’s request on October 9 to delay it to June. Amazon denies the allegations.
• Uber and Lyft: Consumer Reports (CR) measured a 42.4% median gap between the lowest and highest price groups for the same trip. The widely repeated 49.8% figure is the version CR corrected in June. CR did not show that personal data set those fares.
• Delta says its AI tool sees route-level data only. It has not withdrawn its 2024 goal of one day offering each traveler an individual price.
• The law moved fast. The FTC proposed an enforcement policy on August 19. Maryland’s grocery ban took effect October 1. New York’s broader ban awaits the governor. The EU has required disclosure since 2022. China bans hidden data-based price gaps as of April 10, 2026.
• My read: personalization is migrating into discounts and promotions, and the new state bans treat those channels more gently than base prices. How gently depends on conditions that vary by state, and no court has tested them.

Short on time? Jump to the state table, what to do, or the FAQ.

Three terms that get mixed up

Dynamic pricing is legal in every jurisdiction covered here. Almost every dispute below concerns the second and third terms, or whether the first is being used to do something it shouldn’t.

Amazon: the accusation is about the whole market

The FTC and 17 states sued Amazon in 2023. As Washington Monthly summarizes the complaint, it describes three pricing systems:

1. An “anti-discounting” algorithm that matches rivals’ price cuts but never undercuts them.
2. A program called Project Nessie, which allegedly predicted when rivals would follow a price hike and raised prices at those moments. The FTC alleges it produced more than $1 billion in profit from 2016 to 2018.
3. A system that sidelines sellers who price lower elsewhere.

A court let the case proceed in 2024. Washington Monthly reads that ruling as the first to say that tracking and answering rivals’ prices, without collusion, can violate the Sherman Act. The bench trial is scheduled to begin March 29, 2027. California, D.C., and Arizona have separate suits. Everything above is an allegation until a court decides.

The seller side has its own evidence. Economist Leon Musolff studied thousands of Amazon products in 2019 and 2020 and found a cycle: prices jump overnight, then erode as sellers undercut each other. Over time, the cycle pulled competitors’ prices up by more than 11% on average. The tactic was rare then. Repricing vendors now sell it openly, under names like “Yo-Yo Repricing” and “Oscillation Strategy.”

Amazon has a structural stake in higher prices. It keeps roughly 45% of a typical U.S. third-party sale in fees (an Institute for Local Self-Reliance estimate, as reported by Washington Monthly), on about $125 billion of third-party sales in 2023.

Nothing here says Amazon quotes you a higher price than your neighbor. The claim is that the price level itself gets steered.

Uber and Lyft: a widespread, and a cause nobody has proven

Between March and April 2026, Consumer Reports had 174 volunteers price more than 40 routes on both apps, generally within six minutes of each other and often within the same minute. Every one of the 30 virtual routes returned at least two price clusters at least 5% apart. In one Kansas City test, 55 riders on the same Lyft trip saw 29 different prices.

The number everyone quotes is out of date. On June 25, Uber told CR it had found three data-collection errors in volunteers’ screenshots. CR removed those entries. The median gap fell from 49.8% to 42.4%, and the share of discounts CR judged fictitious rose from 10.8% to 12.4%. Several outlets, and a CR petition page I checked, still carry the earlier figures.

Nearly half of upfront prices showed a discount. CR judged 12.4% of advertised discounts fake, meaning the crossed-out “original” price was one almost nobody paid. Uber says crossed-out prices beside labels like “Fares lower than usual” are historical comparisons, not discount claims.

What CR did not claim. It says it is not disputing that the companies avoid personalizing base fares. Its question is whether the gaps come only from market forces. Patents show what the apps could do: Uber’s mention typing speed and phone angle, Lyft’s include a “willingness-to-pay” score. A patent is not proof of practice. Both companies say the only personalization is in promotions, and CR notes neither would list the signals involved.

The companies’ objections. Uber argues that prices change nearly every second, so no two requests are truly the same trip. Lyft pointed to an “observer effect” from dozens of volunteers checking at once. Experts CR consulted dispute both points, since volunteers converged on the same price on nearly every route.

What happened next. On July 7, CR asked the FTC and state attorneys general to investigate. On June 29, the Pennsylvania House passed HB 2512 by 198 to 4. The bill would bar ride-hail fares based on device hardware, installed software, or geolocation. It is now in a Senate committee.

Delta: what it has said, and what it hasn’t

In mid-2025, Delta said about 3% of its domestic fares were priced with an AI system from Fetcherr and that it aimed for 20% by year-end (U.S. News). I found no later public figure.

Three senators wrote to Delta that July. Delta replied that the tool does not use personal data, and lawmakers pressed for specifics again in November 2025. In August 2026, Rep. Frank Pallone wrote to Delta and seven other airlines. Delta’s September 9 reply says only route-level demand and market data go to the tool, and that no fare product Delta has used, tested, or plans uses personal data to set individual prices.

The gap. Delta has not withdrawn the November 2024 investor-day idea of eventually offering a price to “you, the individual.” Its denial covers what the system does today. In August, CEO Ed Bastian said AI could lift profitability by as much as 50%, meaning a margin moving from 10% to 15%. Fetcherr’s CEO has described the technology as using “all the data we can get our hands on,” which Delta’s letters do not address.

Both statements can be true at once. That is why this story keeps going.

Where the law attaches to a price

Original analysis, October 2026. Coverage differs by law; see the sections below. Free to reuse with a link back to this article.

The law, region by region

Status as of October 26, 2026. Several items are unresolved and could change before year-end.

United States: federal

The FTC’s proposed policy statement is the biggest development since the original version of this piece. On August 19, 2026, the Commission voted 2 to 0 to propose an enforcement policy on personalized pricing. It concedes that Congress has not given it power to ban the practice and says it will use Section 5 of the FTC Act instead.

Where shoppers reasonably expect one uniform price, the FTC says sellers should clearly disclose three things: that the price is personalized, the basis for it, and the types of data used. Collecting or using personal data for this without adequate disclosure or consent could be a separate Section 5 violation.

Two limits matter. A policy statement is not a trade rule, so by itself it cannot support monetary relief. And comments closed in September; I found no final version as of today.

Around it, the Senate Judiciary Committee held an August 4 hearing titled “Your Data, Their Profit,” and the House Oversight Committee opened its own inquiry in March.

Enforcement so far is thin. Instacart ended its item-level price tests in December 2025 after CR and Groundwork Collaborative found an average 7% basket variance across 437 shoppers. Its $60 million FTC settlement the same week concerned fees and subscriptions, not the tests. On the collusion side, the DOJ settled with RealPage in November 2025.

United States: states

The AI Laws by State tracker counted 72 bills in 31 states, D.C. and Congress on September 30. Four states had enacted laws.

What the table hides. The three bans do not exempt promotions in the same way, and the differences matter more than the headline.

• Maryland exempts promotional pricing, loyalty and rewards programs, and subscriptions, according to law-firm summaries. One summary adds that the law applies only where personal data is used to raise a price.
• Connecticut carves out voluntary loyalty programs and group discounts, but one law-firm analysis reads loyalty discounts as in scope where a retailer uses data to extend offers that differ among members.
• New Jersey permits loyalty discounts only if enrollment is voluntary, all members get uniform pricing benefits, and the program publicly discloses its benefits and data practices (per a Skadden summary).

So my earlier shorthand, that the fastest-moving laws exempt the channel under scrutiny, is too simple. Flat, disclosed loyalty discounts look safe. An app that targets different riders with different promotions, which is what CR’s findings point toward, may fall inside New Jersey’s and Connecticut’s rules. Whether it does is untested. The FTC’s draft is broader still and reaches individualized discounts and coupons. Quote each statute’s exemption text before relying on any of this.

European Union

The headline deadline moved, and the rules that matter most did not. Regulation (EU) 2026/1744, the Digital Omnibus on AI, entered into force on July 27, 2026. It pushed the AI Act’s stand-alone high-risk obligations from August 2026 to December 2, 2027. Pricing is not an Annex III high-risk category anyway.

Three older rules do the actual work:

• Consumer Rights Directive, Art. 6(1)(ea): since May 28, 2022, traders must tell consumers when a price was personalized by automated decision-making (Directive 2019/2161). That is three years before New York’s banner.
• Price Indication Directive, Art. 6a: an advertised discount must reference the lowest price of the previous 30 days (Directive 98/6/EC). That is the rule CR’s fake-discount finding would run into in Europe.
• AI Act, Art. 5: since February 2025, it bans manipulative systems and those that exploit vulnerabilities such as age, disability, or economic situation, when they cause significant harm (Regulation 2024/1689). Ordinary demand-based pricing is not on that list. The Omnibus left Art. 5 untouched.

GDPR Art. 22 on automated decisions with significant effects is a fourth lever privacy regulators can use.

China

China regulates the mirror image of the Western worry: the platform that charges its loyal users more. The Rules on Pricing Practices for Internet Platforms, issued by the NDRC, SAMR and CAC, took effect on April 10, 2026, for five years. Platforms may not use data or algorithms to set different prices for the same goods under equivalent conditions, based on a consumer’s willingness or ability to pay, preferences or habits, without transparency to the consumer.

This piece does not cover other APAC markets, where I found no pricing-specific rules.

What to do with this

If you run pricing for a retailer or marketplace

1. Keep “dynamic” and “personalized” separate in your architecture and your vendor contracts. They carry different legal exposure.
2. Map every point where personal data can touch a price, including promotion and discount engines. That is where the FTC draft and CR’s findings both land.
3. Build jurisdiction-aware notices: New York’s banner now, Connecticut’s when it takes effect, the EU personalization notice.
4. Keep 30-day price history if you advertise discounts to EU consumers, to back the reference price.
5. Ask your pricing vendor whether its model is shared across competitors in your category. Shared algorithms are the pattern the RealPage case targeted.
6. Test public claims about AI pricing as if a nonprofit will try to falsify them with real transactions. Delta’s and Uber’s disputes both turned on claims outsiders could test.

If you are the shopper

• Compare apps. An NBER study cited by NBC found only 1 in 6 riders checks both Uber and Lyft, and prices differed by 14% on average.
• Treat crossed-out ride prices with suspicion. CR found about one in eight advertised discounts looked fake.
• In New York, a banner on a price means personal data was used to set it.

What could be wrong with this analysis?

• CR’s sample is not representative. CR says so itself, and it did not control for driver supply, pickup estimates or routing. Uber’s timing objection may explain part of the spread.
• Delta may be telling the literal truth. A true denial today and a long-term ambition are compatible.
• The Amazon claims are allegations, and Musolff’s data is from 2019 and 2020.
• Sources disagree on dates. Connecticut is the clearest example. Check statutes before relying on an effective date.
• The promotion-exemption analysis rests on law-firm summaries, which disagree on details (for example, Connecticut’s loyalty scope), not on a line-by-line reading of each statute.
• I work from law-firm summaries and press coverage, not every primary text. I ran no price tests of my own.

FAQ

Is personalized pricing legal in the U.S.?

In most states, yes. No federal law bans it, and the FTC says it lacks authority to ban it outright. What is changing is disclosure and sector bans, as in the table above.

Is New York’s pricing banner actually enforced?

Yes, since November 10, 2025. A federal court rejected the retail lobby’s First Amendment challenge in October 2025, but the Second Circuit appeal was argued on October 19 and could change that.

Does the EU AI Act ban dynamic or personalized pricing?

No. Dynamic pricing is outside the high-risk categories. Personalized pricing is allowed unless it manipulates people or exploits a vulnerability and causes significant harm. Disclosure comes from consumer law, in force since 2022.

Did Uber and Lyft admit to personalizing prices?

They deny personalizing base fares. They say they use personal data for promotions and discounts, which Consumer Reports confirmed in its reporting.

Did Delta use AI to set individual fares?

Delta says no and has said so in writing to Congress twice. Its AI tool does price a portion of its domestic fares, and the company has not withdrawn its longer-term goal of individual offers.

Where are there actual bans today?

Maryland, for food retailers and delivery platforms, since October 1, 2026. Connecticut and New Jersey follow in 2027. China has restricted hidden data-based price gaps for platforms since April 10, 2026.

Will Amazon’s pricing case be decided soon?

Not yet. The FTC’s bench trial is set for March 29, 2027, and a ruling would come after it.

The question has shifted. Regulators no longer ask only whether prices move. They ask whether a price moved because of you, and whether anyone told you. For now, the clearest answer sits in the fine print of promotions and loyalty programs.

About this article

Tom Morgan writes about AI personalization and pricing at AI Personalization Cloud. This piece rests on published documents: court filings, statutes, agency statements, investigations, and law-firm analyses. I ran no price tests of my own; I am not a lawyer, and I read Chinese and EU rules mostly through English-language summaries. No sponsorship, no affiliate links.